Fixed Priority Scheduling

Transcription

1 Fixed Priority Scheduling Giuseppe Lipari Scuola Superiore Sant Anna Pisa January 13, 2011

2 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

3 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

4 The fixed priority scheduling algorithm very simple scheduling algorithm; every task τ i is assigned a fixed priority p i ; the active task with the highest priority is scheduled. Priorities are integer numbers: the higher the number, the higher the priority; In the research literature, sometimes authors use the opposite convention: the lowest the number, the highest the priority. In the following we show some examples, considering periodic tasks, and constant execution time equal to the period.

5 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

6 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

7 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

8 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

9 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

10 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

11 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

12 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

13 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

14 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

15 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

16 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

17 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

18 Example of schedule Consider the following task set: = (2, 6, 6), = (2, 9, 9), τ 3 = (3, 12, 12). Task has priority p 1 = 3 (highest), task has priority p 2 = 2, task τ 3 has priority p 3 = 1 (lowest). τ

19 Another example (non-schedulable) Consider the following task set: = (3, 6, 6), p 1 = 3, = (2, 4, 8), p 2 = 2, τ 3 = (2, 12, 12), p 3 = 1. τ In this case, task τ 3 misses its deadline!

20 Another example (non-schedulable) Consider the following task set: = (3, 6, 6), p 1 = 3, = (2, 4, 8), p 2 = 2, τ 3 = (2, 12, 12), p 3 = 1. τ In this case, task τ 3 misses its deadline!

21 Note Some considerations about the schedule shown before: The response time of the task with the highest priority is minimum and equal to its WCET. The response time of the other tasks depends on the interference of the higher priority tasks; The priority assignment may influence the schedulability of a task.

22 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

23 Priority assignment Given a task set, how to assign priorities? There are two possible objectives: Schedulability (i.e. find the priority assignment that makes all tasks schedulable) Response time (i.e. find the priority assignment that minimize the response time of a subset of tasks). By now we consider the first objective only An optimal priority assignment Opt is such that: If the task set is schedulable with another priority assignment, then it is schedulable with priority assignment Opt. If the task set is not schedulable with Opt, then it is not schedulable by any other assignment.

24 Optimal priority assignment Given a periodic task set with all tasks having deadline equal to the period ( i, D i = T i ), and with all offsets equal to 0 ( i, φ i = 0): The best assignment is the Rate Monotonic assignment Tasks with shorter period have higher priority Given a periodic task set with deadline different from periods, and with all offsets equal to 0 ( i, φ i = 0): The best assignement is the Deadline Monotonic assignment Tasks with shorter relative deadline have higher priority For sporadic tasks, the same rules are valid as for periodic tasks with offsets equal to 0.

25 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

26 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

27 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

28 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

29 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

30 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

31 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

32 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

33 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

34 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

35 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

36 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

37 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

38 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

39 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

40 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

41 Example revised Consider the example shown before with deadline monotonic: = (3, 6, 6), p 1 = 2, = (2, 4, 8), p 2 = 3, τ 3 = (2, 10, 12), p 3 = 1. τ

42 Presence of offsets If instead we consider periodic tasks with offsets, then there is no optimal priority assignment In other words, if a task set T 1 is schedulable by priority O 1 and not schedulable by priority assignment O 2, it may exist another task set T 2 that is schedulable by O 2 and not schedulable by O 1. For example, T 2 may be obtained from T 1 simply changing the offsets!

43 Example of non-optimality with offsets Example: priority to : Changing the offset:

44 Example of non-optimality with offsets Example: priority to : Example: priority to : Changing the offset: Changing the offset:

45 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

46 Analysis Given a task set, how can we guarantee if it is schedulable of not? The first possibility is to simulate the system to check that no deadline is missed; The execution time of every job is set equal to the WCET of the corresponding task; In case of periodic task with no offsets, it is sufficient to simulate the schedule until the hyperperiod (H = lcm i (T i )). In case of offsets, it is sufficient to simulate until 2H +φ max (Leung and Merril). If tasks periods are prime numbers the hyperperiod can be very large!

47 Example Exercise: Compare the hyperperiods of this two task sets: 1 T 1 = 8, T 2 = 12, T 3 = 24; 2 T 1 = 7, T 2 = 12, T 3 = 25. In case of sporadic tasks, we can assume them to arrive at the highest possible rate, so we fall back to the case of periodic tasks with no offsets!

48 Example Exercise: Compare the hyperperiods of this two task sets: 1 T 1 = 8, T 2 = 12, T 3 = 24; 2 T 1 = 7, T 2 = 12, T 3 = 25. In case of sporadic tasks, we can assume them to arrive at the highest possible rate, so we fall back to the case of periodic tasks with no offsets! In case 1, H = 24;

49 Example Exercise: Compare the hyperperiods of this two task sets: 1 T 1 = 8, T 2 = 12, T 3 = 24; 2 T 1 = 7, T 2 = 12, T 3 = 25. In case of sporadic tasks, we can assume them to arrive at the highest possible rate, so we fall back to the case of periodic tasks with no offsets! In case 1, H = 24; In case 2, H = 2100!

50 Utilization analysis In many cases it is useful to have a very simple test to see if the task set is schedulable. A sufficient test is based on the Utilization bound: Definition The utilization least upper bound for scheduling algorithm A is the smallest possible utilization U lub such that, for any task set T, if the task set s utilization U is not greater than U lub (U U lub ), then the task set is schedulable by algorithm A. U U lub t

51 Utilization bound for RM Theorem (Liu and Layland, 1973) Consider n periodic (or sporadic) tasks with relative deadline equal to periods, whose priorities are assigned in Rate Monotonic order. Then, U lub = n(2 1/n 1) U lub is a decreasing function of n; For large n: U lub 0.69 n U lub n U lub

52 Schedulability test Therefore the schedulability test consist in: Compute U = n C i i=1 T i ; if U U lub, the task set is schedulable; if U > 1 the task set is not schedulable; if U lub < U 1, the task set may or may not be schedulable;

53 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

54 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

55 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

56 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

57 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

58 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

59 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

60 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

61 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

62 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

63 Example Example in which we show that for 3 tasks, if U < U lub, the system is schedulable. = (2, 8), = (3, 12),τ 3 = (4, 16); U = 0.75 < U lub = 0.77 τ

64 Example 2 By increasing the computation time of task τ 3, the system may still be schedulable... = (2, 8), = (3, 12),τ 3 = (5, 16); U = 0.81 > U lub = 0.77 τ

65 Utilization bound for DM If relative deadlines are less than or equal to periods, instead of considering U = n C i i=1 T i, we can consider: U = n i=1 C i D i Then the test is the same as the one for RM (or DM), except that we must use U instead of U.

66 Pessimism The bound is very pessimistic: most of the times, a task set with U > U lub is schedulable by RM. A particular case is when tasks have periods that are harmonic: A task set is harmonic if, for every two tasks τ i, tau j, either P i is multiple of P j or P j is multiple of P i. For a harmonic task set, the utilization bound is U lub = 1. In other words, Rate Monotonic is an optimal algoritm for harmonic task sets.

67 Example of harmonic task set = (3, 6), = (3, 12), τ 3 = (6, 24); U = 1 ; τ

68 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

69 Response time analysis A necessary and sufficient test is obtained by computing the worst-case response time (WCRT) for every task. For every task τ i : Compute the WCRT R i for task τ i ; If R i D i, then the task is schedulable; else, the task is not schedulable; we can also show the situation that make task τ i miss its deadline! To compute the WCRT, we do not need to do any assumption on the priority assignment. The algorithm described in the next slides is valid for an arbitrary priority assignment. The algorithm assumes periodic tasks with no offsets, or sporadic tasks.

70 Response time analysis - II The critical instant for a set of periodic real-time tasks, with offset equal to 0, or for sporadic tasks, is when all jobs start at the same time. Theorem (Liu and Layland, 1973) The WCRT for a task corresponds to the response time of the job activated at the critical instant. To compute the WCRT of task τ i : We have to consider its computation time and the computation time of the higher priority tasks (interference); higher priority tasks can preempt task τ i, and increment its response time.

71 Response time analysis - III Suppose tasks are ordered by decreasing priority. Therefore, i < j prio i > prio j. Given a task τ i, let R (k) i be the WCRT computed at step k. R (k) i R (k) i i 1 i = C i + R (0) R (k) j=1 j=1 C j i 1 (k 1) R i i = C i + C j T j The iteration stops when: = R (k+1) i or > D i (non schedulable);

72 Example Consider the following task set: = (2, 5), = (2, 9), τ 3 = (5, 20); U = R (k) i 1 i = C i + j=1 (k 1) R i T j C j R (0) 3 = C C C 2 = 9 τ

73 Example Consider the following task set: = (2, 5), = (2, 9), τ 3 = (5, 20); U = R (k) i 1 i = C i + j=1 (k 1) R i T j C j R (0) 3 = C C C 2 = 9 R (1) 3 = C C C 2 = 11 τ

74 Example Consider the following task set: = (2, 5), = (2, 9), τ 3 = (5, 20); U = R (k) i 1 i = C i + j=1 (k 1) R i T j C j R (0) 3 = C C C 2 = 9 R (1) 3 = C C C 2 = 11 R (2) 3 = C C C 2 = 15 τ

75 Example Consider the following task set: = (2, 5), = (2, 9), τ 3 = (5, 20); U = R (k) i 1 i = C i + j=1 (k 1) R i T j C j R (0) 3 = C C C 2 = 9 R (1) 3 = C C C 2 = 11 R (2) 3 = C C C 2 = 15 R (3) 3 = C C C 2 = 15 = R (2) 3 τ

76 Another example with DM The method is valid for different priority assignments and deadlines different from periods = (1, 4, 4), p 1 = 3, = (4, 6, 15), p 2 = 2, τ 3 = (3, 10, 10), p 3 = 1; U = 0.72 R (0) 3 = C C C 2 = 8 R (k) i 1 (k 1) R i i = C i + C j T j=1 j τ

77 Another example with DM The method is valid for different priority assignments and deadlines different from periods = (1, 4, 4), p 1 = 3, = (4, 6, 15), p 2 = 2, τ 3 = (3, 10, 10), p 3 = 1; U = 0.72 R (0) 3 = C C C 2 = 8 R (1) 3 = C C C 2 = 9 R (k) i 1 (k 1) R i i = C i + C j T j=1 j τ

78 Another example with DM The method is valid for different priority assignments and deadlines different from periods = (1, 4, 4), p 1 = 3, = (4, 6, 15), p 2 = 2, τ 3 = (3, 10, 10), p 3 = 1; U = 0.72 R (0) 3 = C C C 2 = 8 R (1) 3 = C C C 2 = 9 R (2) 3 = C C C 2 = 10 R (k) i 1 (k 1) R i i = C i + C j T j=1 j τ

79 Another example with DM The method is valid for different priority assignments and deadlines different from periods = (1, 4, 4), p 1 = 3, = (4, 6, 15), p 2 = 2, τ 3 = (3, 10, 10), p 3 = 1; U = 0.72 R (0) 3 = C C C 2 = 8 R (1) 3 = C C C 2 = 9 R (2) 3 = C C C 2 = 10 R (k) i 1 (k 1) R i i = C i + C j T j=1 j R (3) 3 = C C C 2 = 10 = R (2) 3 τ

80 Considerations The response time analysis is an efficient algorithm In the worst case, the number of steps N for the algorithm to converge is exponential It depends on the total number of jobs of higher priority tasks that may be contained in the interval [0, D i ]: i 1 N j=1 Di If s is the minimum granularity of the time, then in the worst case N = D i s ; However, such worst case is very rare: usually, the number of steps is low. T j

81 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

82 Considerations on WCET The response time analysis is a necessary and sufficient test for fixed priority. However, the result is very sensitive to the value of the WCET. If we are wrong in estimating the WCET (and for example we put a value that is too low), the actual system may be not schedulable. The value of the response time is not helpful: even if the response time is well below the deadline, a small increase in the WCET of a higher priority task makes the response time jump; We may see the problem as a sensitivity analysis problem: we have a function R i = f i (C 1, T 1, C 2, T 2,...,C i 1, T i 1, C i ) that is non-continuous.

83 Example of discontinuity Let s consider again the example done before; we increment the computation time of of 0.1. τ

84 Example of discontinuity Let s consider again the example done before; we increment the computation time of of 0.1. τ R 3 =

85 Singularities The response time of a task τ i is the first time at which all tasks,...,τ i have completed; At this point, either a lower priority task τ j (p j < p i ) is executed or the system becoms idle or it coincides with the arrival time of a higher priority task. In the last case, such an instant is also called i-level singularity point. In the previous example, time 12 is a 3-level singularity point, because: 1 task τ 3 has just finished; 2 and task ha just been activated; A singularity is a dangerous point!

86 Sensitivity on WCETs A rule of thumb is to increase the WCET by a certain percentage before doing the analysis. If the task set is still feasible, be are more confident about the schedulability of the original system. There are analytical methods for computing the amount of variation that it is possible to allow to a task s WCET without compromising the schedulability

87 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

88 A different analysis approach Definition of workload for task τ i : W i (t) = i t j=1 T j C j The workload is the amount of work that the set of tasks {,...,τ i } requests in [0, t] Example: = (2, 4), = (4, 15): W 2 (10) = 2+ 4 = 6+4 =

89 Workload function The workload function for the previous example = (2, 4), = (4, 15):

90 Main theorem Theorem (Lehokzcy 1987) Let P i = { j < i, k, kt j D i kt j } {D i }. Then, task τ i is schedulable if and only if t P i, W i (t) t Set P i is the set of time instants that are multiple of some period of some task τ j with higher priority than τ i, plus the deadline of task τ i (they are potential singularity points) In other words, the theorem says that, if the workload is less than t for any of the points in P i, then τ i is schedulable Later, Bini simplified the computation of the points in set P i

91 Example with 4 tasks = (2, 4), = (4, 15), τ 3 = (4, 30), τ 4 = (4, 60) Task τ 4 is schedulable, because W 4 (56) = 56 and W 4 (60) = 58 < 60 (see schedule on fp_schedule_1.0_ex4.ods)

92 Sensitivity analysis Proposed by Bini and Buttazzo, 2005 Let us rewrite the equations for the workload: t P i i t j=1 T j C j t If we consider the C j as variables, we have a set of linear inequalities in OR mode each inequality defines a plane in the space R i of variables C 1,...,Ci the result is a concave hyper-solid in that space

93 Example with two tasks = (x, 4), = (y, 15) P = {4, 8, 12, 15} C 1 + C 2 4 2C 1 + C 2 8 3C 1 + C C 1 + C 2 15

94 Graphical representation In the R 2 space: C2 C 1

95 Simplifying non-useful constraints C2 The cross represent a (possible) pair of values for (C 1, C 2 ). The cross must stay always inside the subspace C 1

96 Sensitivity Distance from a constraint represents how much we can increase (C 1, C 2 ) without exiting from the space or how much we must decrease C 1 or C 2 to enter in the space In the example before: starting from C 1 = 1 and C 2 = 8 we can increase C 1 of the following: 3(1+ ) = 1 3 Exercise: verify schedulability of with C 1 = and C 2 = 8 by computing its response time

97 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

98 Summary of schedulability tests for FP Utilization bound test: depends on the number of tasks; for large n, U lub = 0.69; only sufficient; O(n) complexity; Response time analysis: necessary and sufficient test for periodic tasks with arbitrary deadlines and with no offset complexity: high (pseudo-polynomial); Hyperplane analysis necessary and sufficient test for periodic tasks with arbitrary deadlines and with no offset complexity: high (pseudo-polynomial); allows to perform sensitivity analysis

99 Response time analysis - extensions Consider offsets Arbitrary patterns of arrivals. Burst, quasi periodic, etc.

100 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

101 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

102 Esercizio Dato il seguente task set: Task C i D i T i τ τ Calcolare il tempo di risposta dei vari task nell ipotesi che le priorità siano assegnate con RM o con DM. Risposta: Nel caso di RM, R( ) = 1 R( ) = 3 R(τ 3 ) = 7 R(τ 4 ) = 18

103 Esercizio Dato il seguente task set: Task C i D i T i τ τ Calcolare il tempo di risposta dei vari task nell ipotesi che le priorità siano assegnate con RM o con DM. Risposta: Nel caso di RM, Nel caso di DM, R( ) = 1 R( ) = 3 R(τ 3 ) = 7 R(τ 4 ) = 18 R( ) = 1 R( ) = 7 R(τ 3 ) = 4 R(τ 4 ) = 18

104 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

105 Esercizio Consideriamo il seguente task non periodico: Se j è pari, allora a 1,j = 8 j 2 ; Se j è dispari, allora a 1,j = 3+8 j 2 ; In ogni caso, c 1,j = 2; La priorità del task è p 1 = 3. Nel sistema, consideriamo anche i task periodici = (2, 12, 12) e τ 3 = (3, 16, 16), con priorità p 2 = 2 e p 3 = 1. Calcolare il tempo di risposta dei task e τ 3.

106 Soluzione - I Il pattern di arrivo del task è il seguente: Il task è ad alta priorità, quindi il suo tempo di risposta è pari a 2. In che modo questo task interferisce con gli altri due task a bassa priorità

107 Soluzione - II Bisogna estendere la formula del calcolo del tempo di risposta. La generalizzazione è la seguente: R (k) i 1 i = C i + Nist j (R (k 1) i )C j j=1 dove Nist j (t) rappresenta il numero di istanze del task τ j che arrivano nell intervallo [0, t). Se il task τ j è periodico, allora Nist j (t) = t. Tj Nel caso invece del task (che non èperiodico): t max(0, t 3) Nist 1 (t) = Il primo termine tiene conto delle istanze con j pari, mentre il secondo termine tiene conto delle istanze con j dispari.

108 Soluzione - III Applicando la formula per calcolare il tempo di risposta del task : R (0) 2 = 2+2 = 4 R (1) 2 = = 6 R (2) 2 = = 6

109 Soluzione - III Applicando la formula per calcolare il tempo di risposta del task : Per il task τ 3 : R (0) 2 = 2+2 = 4 R (1) 2 = = 6 R (2) 2 = = 6 R (0) 3 = = 7 R (1) 3 = = 9 R (2) 3 = = 11 R (3) 3 = = 11

110 Soluzione - IV (schedulazione) Schedulazione risultante: τ

111 Outline 1 Fixed priority 2 Priority assignment 3 Scheduling analysis 4 A necessary and sufficient test 5 Sensitivity 6 Hyperplane analysis 7 Conclusions 8 Esercizi Calcolo del tempo di risposta Calcolo del tempo di risposta con aperiodici Hyperplane analysis

112 Esercizio sulla sensitivity Dato il seguente insieme di task: = (2, 5), = (3, 12) Vedere se il sistema è schedulabile con l analisi Hyperplanes Calcolare di quando può aumentare (o di quanto si può diminuire) il tempo di calcolo di per farlo rimanere (diventare) schedulabile Calcolare di quanto si può diminuire la potenza del processore mantenendo il sistema schedulabile

113 Soluzione Le equazioni da considerare sono: C 1 + C 2 5 2C 1 + C C 1 + C 2 12 Tutte verificate per C 1 = 2 e C 2 = 3 Fissando C 1, si ha: C 2 3 C 2 6 C 2 6 Ricordandoci che sono in OR, la soluzione è C 2 6, quindi possiamo aumentare C 2 di 3 mantenendo il sistema schedulabile

114 Soluzione 2 Se il processore ha velocità variabile, le equazioni possono essere riscritte come: E nel punto considerato: αc 1 +αc 2 5 2αC 1 +αc αC 1 +αc 2 12 α 1 7α 10 9α 12 Quindi, α = , e possiamo rallentare il processore (cioé incrementare i tempi di calcolo) del 43% circa.